[Top] [All Lists]

Re: [PATCH 2/2] xfs: do not call xfs_bdstrat_cb in xfs_buf_iodone_callba

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 2/2] xfs: do not call xfs_bdstrat_cb in xfs_buf_iodone_callbacks
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Wed, 4 Jul 2012 11:16:50 -0400
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <20120704083234.GG19223@dastard>
References: <20120702100003.960640484@xxxxxxxxxxxxxxxxxxxxxx> <20120702100034.921366796@xxxxxxxxxxxxxxxxxxxxxx> <20120703002857.GY19223@dastard> <20120703160531.GA855@xxxxxxxxxxxxx> <20120703232923.GC19223@dastard> <20120704055723.GB27500@xxxxxxxxxxxxx> <20120704083234.GG19223@dastard>
User-agent: Mutt/1.5.21 (2010-09-15)
On Wed, Jul 04, 2012 at 06:32:34PM +1000, Dave Chinner wrote:
> IMO, using uncached buffers in transactions is dangerous because
> concurrent transactions can't find buffers uncached buffers at the
> same address and so such buffers need higher level synchronisation
> interfaces (as the superblock has).
> I'm happy to leave the superblock as uncached and work around it in
> some way, but it just seems wrong to me to have one buffer behaves
> differently to all the cached and other uncached buffers in the
> system. It doesn't sit well in my mind to do that when the problem
> simply goes away if we make the superblock a cached buffer again....

The superblock buffer always has been more special than others, remember
the old FS_MANAGED flag before we went for uncached buffers?

If you want to move it back to normal buffers how do you want to look
it up, given that IIRC we read it before setting up the perag buffers.

<Prev in Thread] Current Thread [Next in Thread>